This document explains how to install and configure wandb.
Step 1: Installation of wandb
This install of wandb is included when installing the SubVortex subnet (see the guide for installation details)
You can check wandb is installing by executing in SubVortex directory
pip show wandb
You have to see something like
Name: wandb
Version: 0.16.5
Summary: A CLI and library for interacting with the Weights & Biases API.
Home-page:
Author:
Author-email: Weights & Biases <[email protected]>
License: MIT License
Step 2: Obtain Your API Key
- Log in to your Weights & Biases account through your web browser.
- Go to your account settings, usually accessible from the top right corner under your profile.
- Find the section labeled "API keys".
- Copy your API key. It's a long string of characters unique to your account.
Step 3: Setting Up the API Key in Ubuntu
Log into wandb by executing
wandb login
You are going to be asked to provide your api key
wandb: Logging into wandb.ai. (Learn how to deploy a W&B server locally: https://wandb.me/wandb-server)
wandb: You can find your API key in your browser here: https://wandb.ai/authorize
wandb: Paste an API key from your profile and hit enter, or press ctrl+c to quit:
The default configuration is enough to have a good user experience so there is no real need to update it.
The default configuration will create a maximum of 1 run (active) containing 360 steps of data. We chose 360, which corresponds to an epoch, and we believe it is sufficient to understand the trend and adjust the subtensor accordingly.
IMPORTAMT
For now, we choose to have one active run per validator in order to stay in the free plan as this dahsboard will be replaced by our own Frontend at some point.
Options
--wandb.off
- turn off wandb. Default false.--wandb.project_name
- The name of the project where you are sending the new runs. Default is subvortex-team for mainnet and test-subvortex-team for testnet.--wandb.entity
- An entity is a username or team name where youre sending runs. Default is eclipsevortext.--wandb.offline
- Runs wandb in offline mode. Default false.--wandb.run_step_length
- How many steps before we rollover to a new run. Default 360.
To enhance user experience, we have decided to name each run using the format validator-<VALIDATOR_UID>-<COUNT>
. This format allows users to easily identify the runs of the validator they wish to access. The count will increment with each run until it reaches 10000, at which point it will reset.
For miner, wandb is not needed so nothing to do here.
To access the wandb UI to get statistics about the miners, you can click on this folling links and choose the validator run you want
- mainnet -
subvortex-team
- testnet -
test-subvortex-team
The line plot provides valuable insights, including:
- The performance of the best miner (UID) over time.
- The step time evolution, representing the time taken for the validator to test all subtensors and compute the final score.
The table display the list of miners with the following informations
UID
- UID of the miner.Version
- Version of the miner.Country
- Geographical location of the subtensor.Score
- Last final score of the subtensor.Availability
- Last availability score of the subtensor.Reliability
- Last reliability score of the subtensor.Latency
- Last latency score of the subtensor.Distribution
- Last distribution score of the subtensor.
From that table, you can get a quick overview of how you are competing with others. You can sort the different columns to prioritize the desired information. We are still investigating if filtering is possible with Wandb.
These historgras gives you the number of subtensors per country
Verified Miners Distribution
- number of verified subtensor per country - A subtensor is considered verified when 1) the miner is up and running 2) the subtensor is up and running and 3) there is only one miner associated to the subtensor. This histogram is used to check the distribution score your miner is getting.Miners Distribution
- number of subtensor (verified or not) per country - This histogram can be used to determine which countries do not have any subtensors yet, in order to set up one and attain the maximum associated score.
The line plot gives you the an evolution of the score through time. There is one line plot per score (final, availability, reliability, latency and distribution) and per miner.
With these scores, you can understand how your miners compete with others and make the appropriate adjustments of your choice.
The line plot provides valuable insights, including:
- The process time evolution, representing the time taken for the validator to get the current block from the subtensor. This time is a rwa time and does not include the tolerance based on the distance between the validator and the subtensor.
The search bar can be use to display only the scores related to your miner. To do so, just seize in the search input _score.<UID
and replace UID by your miner ID. You are going to see something like below
This wandb UI is built as a temporary solution to provide miners and validators with tools and data as quick as possible. A more professional and robust front-end website will developped in the future to further enhance the user experience.
SubVortex will be actively engaging with our users to gauge their needs and preferences. Based on their feedback, we will implement strategic changes to the provided tools. We will continue to ensure all users have the necessary information needed to make informed decisions to optimize their performance.
At that time, we could not guarantee the sorting to work every time.
To resolve your issue, either drag and drop manually the sections or click on the settings on the right on the Search panel with regex
input
Then, click to the locker to unlock the Prtected settings
section
Choose the empty option in the Organize panels into sections by metric name
section, and then re-select the initial option. Ensure that the selection remains the same as before. While performing this action, you might observe changes on the page behind the modal.
Finally, click on Done
to see the sections sorted as expected